Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Semi-Automated Data Annotation in Multisensor Datasets for Autonomous Vehicle Testing

Andrii Gamalii, Daniel Górniak, Robert Nowak, Bartłomiej Olber, Krystian Radlak, Jakub Winter

TL;DR

The paper presents a semi-automated, human-in-the-loop data annotation pipeline (DARTS) for building a large-scale, multimodal driving dataset tailored to Polish road conditions. It combines AI-based 3D object detection with domain adaptation, anonymization, and iterative retraining, all orchestrated via Apache Airflow and integrated with Segments.ai for human verification. A novel Correction Acceleration Ratio (CAR) metric quantifies annotation-time savings, and a PostgreSQL-based schema extends nuScenes to track provenance and validation. Empirical results show substantial reductions in manual labeling time (e.g., CAR up to 0.93 on Zenseact) and demonstrate the pipeline’s effectiveness in producing scalable, GDPR-compliant annotations suitable for national autonomous driving data initiatives.

Abstract

This report presents the design and implementation of a semi-automated data annotation pipeline developed within the DARTS project, whose goal is to create a large-scale, multimodal dataset of driving scenarios recorded in Polish conditions. Manual annotation of such heterogeneous data is both costly and time-consuming. To address this challenge, the proposed solution adopts a human-in-the-loop approach that combines artificial intelligence with human expertise to reduce annotation cost and duration. The system automatically generates initial annotations, enables iterative model retraining, and incorporates data anonymization and domain adaptation techniques. At its core, the tool relies on 3D object detection algorithms to produce preliminary annotations. Overall, the developed tools and methodology result in substantial time savings while ensuring consistent, high-quality annotations across different sensor modalities. The solution directly supports the DARTS project by accelerating the preparation of large annotated dataset in the project's standardized format, strengthening the technological base for autonomous vehicle research in Poland.

Semi-Automated Data Annotation in Multisensor Datasets for Autonomous Vehicle Testing

TL;DR

The paper presents a semi-automated, human-in-the-loop data annotation pipeline (DARTS) for building a large-scale, multimodal driving dataset tailored to Polish road conditions. It combines AI-based 3D object detection with domain adaptation, anonymization, and iterative retraining, all orchestrated via Apache Airflow and integrated with Segments.ai for human verification. A novel Correction Acceleration Ratio (CAR) metric quantifies annotation-time savings, and a PostgreSQL-based schema extends nuScenes to track provenance and validation. Empirical results show substantial reductions in manual labeling time (e.g., CAR up to 0.93 on Zenseact) and demonstrate the pipeline’s effectiveness in producing scalable, GDPR-compliant annotations suitable for national autonomous driving data initiatives.

Abstract

This report presents the design and implementation of a semi-automated data annotation pipeline developed within the DARTS project, whose goal is to create a large-scale, multimodal dataset of driving scenarios recorded in Polish conditions. Manual annotation of such heterogeneous data is both costly and time-consuming. To address this challenge, the proposed solution adopts a human-in-the-loop approach that combines artificial intelligence with human expertise to reduce annotation cost and duration. The system automatically generates initial annotations, enables iterative model retraining, and incorporates data anonymization and domain adaptation techniques. At its core, the tool relies on 3D object detection algorithms to produce preliminary annotations. Overall, the developed tools and methodology result in substantial time savings while ensuring consistent, high-quality annotations across different sensor modalities. The solution directly supports the DARTS project by accelerating the preparation of large annotated dataset in the project's standardized format, strengthening the technological base for autonomous vehicle research in Poland.
Paper Structure (18 sections, 3 equations, 4 figures, 2 tables)

This paper contains 18 sections, 3 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Scope of the Task
  3. System Design
  4. System overview
  5. System workflow and orchestration overview
  6. Modules Implementation
  7. Annotation Generator
  8. Anonymization
  9. Data Preprocessing
  10. Database
  11. MOT Tracking
  12. Segments Toolkit
  13. Darts Utils
  14. Evaluation and Results
  15. Motivation for a New Metric
  16. ...and 3 more sections

Figures (4)

  • Figure 1: System context diagram
  • Figure 2: Container diagram
  • Figure 3: Sequence diagram presenting DARTS pipeline
  • Figure 4: Generated annotations imported into Segments.ai